Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Subspace clustering via good neighbors

Yang, Jufeng, Liang, Jie, Wang, Kai, Rosin, Paul L. ORCID: and Yang, Ming-Hsuan 2020. Subspace clustering via good neighbors. IEEE Transactions on Pattern Analysis and Machine Intelligence 42 (6) , pp. 1537-1544. 10.1109/TPAMI.2019.2913863

[thumbnail of clustering-neighbors-postprint.pdf]
PDF - Accepted Post-Print Version
Download (1MB) | Preview


Finding the informative clusters of a high-dimensional dataset is at the core of numerous applications in computer vision, where spectral based subspace clustering algorithm is arguably the most widely-studied methods due to its empirical performance and provable guarantees under various assumptions. It is well-known that sparsity and connectivity of the affinity graph play important rules for effective subspace clustering. However, it is difficult to simultaneously optimize both factors due to their conflicting nature, and most existing methods are designed to deal with only one factor. In this paper, we propose an algorithm to optimize both sparsity and connectivity by finding good neighbors which induce key connections among samples within a subspace. First, an initial coefficient matrix is generated from the input dataset. For each sample, we find its good neighbors which not only have large coefficients but are strongly connected to each other. We reassign the coefficients of good neighbors and eliminate other entries to generate a new coefficient matrix, which can be used by spectral clustering methods. Experiments on five benchmark datasets show that the proposed algorithm performs favorably against the state-of-the-art methods in terms of accuracy with a negligible increase in speed.

Item Type: Article
Date Type: Publication
Status: Published
Schools: Computer Science & Informatics
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
ISSN: 0162-8828
Date of First Compliant Deposit: 4 June 2019
Date of Acceptance: 8 April 2019
Last Modified: 06 Nov 2023 21:48

Citation Data

Cited 28 times in Scopus. View in Scopus. Powered By Scopus® Data

Actions (repository staff only)

Edit Item Edit Item


Downloads per month over past year

View more statistics